Add prefix syntax

[klmr]

The currently supported syntax is

hello = flask$route('/<name>') %@% function (name)
    sprintf('Hello %s', name)

Ideally, the following syntax should be supported:

flask$route('/<name>') %@%
hello = function (name)
    sprintf('Hello %s', name)

Unfortunately, I’m not sure that this is workable with the R syntax rules. The following at least works:

decorator %@%
hello = function (name)
    sprintf('Hello %s', name)

To make this work, all that’s necessary is to provide %@%<-. However, when applying this to a decorator that’s not a simple identifier, R throws the following error:

target of assignment expands to non-language object

The reason for this is logical; R allows assignments of the form f(arguments) = value, provided that the functionf<- exists. However, this only works if the first argument to f is a simple identifier, since R will ultimately re-assign an object to that first argument (see R’s subset assignment documentation). So f('test') = value cannot be made to work (that despite the fact that the assignment 'test' = value is valid in R!).

Making this work may therefore require (backwards-compatible) changes to the R parser. Good luck with that.

[klmr]

Potential solution to the problem with parametrised decorators:

R seems to allow the following syntax:

logged['log.txt'] %@%
echo = function (msg)
    message(msg)

… i.e. using brackets instead of a function call, and overloading [.decorator and [<-.decorator appropriately. I just need to figure out how exactly that function is being called.

[klmr]

To follow up on this, the above is parsed as

─ =
  ├─ %@%
  │  ├─ []
  │  │  ├─ logged
  │  │  └─ "log.txt"
  │  └─ echo
  └─ function (msg = )
     └─ message
        └─ msg

This leads to the following nested assignment implementation:

`*tmp*` <- logged
logged <- `[<-`(`*tmp*`,
                'log.txt',
                value = `%@%<-`(`[`(`*tmp*`, 'log.txt'),
                                echo,
                                value = function (msg) message(msg)))
rm(`*tmp*`)

In other words, the implementation needs to provide

• [<-.decorator,
• [.decorator and
• %@%<-.decorator

with appropriate semantics. Furthermore, %@%.decorator is needed anyway to implement the existing decorator syntax directly in front of function (…) … (still required for anonymous functions).